DI

SC

US

SI

ON

P

AP

ER

S

ER

IE

S

Forschungsinstitut zur Zukunft der ArbeitInstitute for the Study of Labor

Cognitive, Non-Cognitive Skills and Gender Wage Gaps:Evidence from Linked Employer-Employee Datain Bangladesh

IZA DP No. 9132

June 2015

Christophe J. NordmanLeopold R. SarrSmriti Sharma

Cognitive, Non-Cognitive Skills and

Gender Wage Gaps: Evidence from Linked Employer-Employee Data in Bangladesh

Christophe J. Nordman DIAL (IRD & University Paris-Dauphine) and IZA

Leopold R. Sarr

World Bank

Smriti Sharma UNU-WIDER

Discussion Paper No. 9132 June 2015

IZA

P.O. Box 7240 53072 Bonn

Germany

Phone: +49-228-3894-0 Fax: +49-228-3894-180

E-mail: iza@iza.org

Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

IZA Discussion Paper No. 9132 June 2015

ABSTRACT

Cognitive, Non-Cognitive Skills and Gender Wage Gaps: Evidence from Linked Employer-Employee Data in Bangladesh* We use a first-hand linked employer-employee dataset representing the formal sector of Bangladesh to explain gender wage gaps by the inclusion of measures of cognitive skills and personality traits. Our results show that while cognitive skills are important in determining mean wages, personality traits have little explanatory power. However, quantile regressions indicate that personality traits do matter in certain parts of the conditional wage distribution, especially for wages of females. Cognitive skills as measured by reading and numeracy also confer different benefits across the wage distribution to females and males respectively. Quantile decompositions indicate that these skills and traits reduce the unexplained gender gap, mainly in the upper parts of the wage distribution. Finally, results suggest that employers place greater consideration on observables such as academic background and prior work experience, and may also make assumptions about the existence of sex-specific skills of their workers, which could then widen the within-firm gender wage gap. JEL Classification: J16, J24, J31, J71, C21, O12 Keywords: gender wage gap, cognitive skills, personality traits, matched worker-firm data,

quantile decompositions, Bangladesh Corresponding author: Christophe J. Nordman IRD, DIAL 4 rue d’Enghien 75010 Paris France E-mail: nordman@dial.prd.fr

* This paper has benefited from comments received from Paul Glewwe, Prakarsh Singh, Saurabh Singhal and participants at IZA/OECD/World Bank Workshop on Cognitive and Non-Cognitive Skills and Economic Development 2014, 10th Annual Conference on Economic Growth and Development, Indian Statistical Institute, Royal Economic Society Conference 2015, Midwest International Economic Development Conference 2015 and the 10th IZA/World Bank Conference on Employment and Development. The usual disclaimer applies.

1 Introduction

Non-cognitive skills or personality traits have recently received significant attention as

determinants of labour market performance. In fact, these non-cognitive traits, referring

to qualities such as motivation, leadership, self-esteem, social skills, etc., have in some

cases been shown to be at least as important as cognitive skills for wages and employment

prospects (e.g., Heckman et al., 2006; Lindqvist and Vestman, 2011).1 Theoretically,

personality traits can have both direct and indirect effects on productivity (Borghans

et al., 2008). They can affect productivity directly by being considered as part of an

individual’s set of endowments, or serve as incentive-enhancing preferences (Bowles et

al., 2001). Additionally, they can indirectly affect productivity, for instance, through

effects on occupational choice (Cobb-Clark and Tan, 2011) and educational attainment

(Heckman et al., 2011).

There is a growing body of literature that has explored gender differences in per-

sonality traits as potential alternative explanations of the gender wage gap.2 However,

the existing evidence is based predominantly on developed economies with results indi-

cating considerable variation in the contribution of these traits to the wage gaps. For

instance, Mueller and Plug (2006) find that 3 percent of the gender wage gap in U.S. is

explained by differences in personality traits (measured by the Big Five). On the other

hand, Fortin (2008), on U.S. workers also reports that 8 percent of the gender wage gap

is explained by differences in non-cognitive traits such as importance of money/work and

importance of people/family. A similar magnitude has been documented for Russia (Se-

mykina and Linz, 2007), while for Germany the effects are relatively minor (Braakmann,

2009). Using Australian data, Cobb-Clark and Tan (2011) find that men’s and women’s

non-cognitive skills significantly influence sorting into occupations although the nature

of this relationship varies across gender.

In this paper, our objective is to explain gender wage gaps in the formal sector of

Bangladesh as a function of gender differences in cognitive skills and personality traits,

over and above the standard variables included in Mincerian wage regressions. Gen-

der disparities heavily characterize the Bangladeshi labour market. The increase in the

proportion of females in the labour force, from 26 percent in 2002-03 to 36 percent in

2010 (Bangladesh Bureau of Statistics, 2011), has been largely on account of the rapidly

1Heckman et al. (2006) show that if an individual moves from the 25th percentile to the 75th percentilein the distribution of non-cognitive skills, wages at age 30 improve by about 10 percent for males, and bymore than 30 percent for females. In comparison, a similar movement in the cognitive skill distributionleads to a 20 percent wage increase for males and to 30 percent increase for females.

2Further, a large experimental literature shows that men and women tend to differ in behaviouraltraits such as competitiveness (e.g., Niederle and Vesterlund, 2007), risk aversion (e.g., Croson andGneezy, 2009) and willingness to negotiate (e.g., Babcock and Laschever, 2003), factors that can partlyexplain gender differences in outcomes such as job entry (Flory et al., 2015) and wages (Card et al.,2013).

2

growing ready-made garment sector, where females constitute 80 percent of factory work-

ers and approximately 15 percent of women aged 16-30 work in this sector (see Heath

and Mobarak, 2015 and references therein). In terms of wages, Kapsos (2008) finds that

women in the non-agricultural sector earn 21 percent less per hour than men while Ahmed

and Maitra (2010) find that gender wage gaps are substantially higher in urban areas.

While the literature on estimating gender wage gaps in developing and transition coun-

tries is fairly large (e.g., Appleton et al., 1999; Chi and Li, 2008; Nordman and Roubaud,

2009; Nordman et al., 2011), evidence documenting the influence of cognitive and non-

cognitive skills on gender wage gaps in a developing country context is scarce, primarily

due to data limitations. With new data that allow us to identify these skills and traits

for Bangladesh, to the best of our knowledge, our paper is among the first to contribute

to this line of research examining the importance of cognitive and non-cognitive skills in

explaining gender wage gaps in a developing country.3

Further, since looking at gender gaps at the means of men’s and women’s wages may

only reveal part of the prevailing gender inequalities, we also conduct a distributional

analysis of wage gaps. This allows us to analyze how cognitive skills and personality

traits are valued at different points of the wage distribution. Quantile regression based

decomposition techniques, that decompose wage gaps into explained and unexplained

components at various points of the wage distribution, document mostly a “glass ceiling

effect”, i.e. the gender wage gap is increasing at the upper end of the wage distribution,

for developed economies and some developing countries such as Morocco (Albrecht et

al., 2003; Jellal et al., 2003; Nordman and Wolff, 2009a, 2009b). On the other hand, in

Asian developing countries such as India, China, Vietnam and Bangladesh, larger wage

gaps have been observed at the lower tails of the earnings distribution, i.e. the “sticky

floor” phenomenon (Pham and Reilly, 2007; Khanna, 2012; Chi and Li, 2008; Ahmed

and Maitra, 2011). Carrillo et al. (2014) based on their examination of gender wage gaps

in twelve Latin American countries find that poorer and more unequal countries exhibit

sticky floors whereas glass ceilings characterize richer and less unequal ones.

Since our data are collected at the enterprise level and contain information about

both firms and employees, we make use of the linked employer-employee nature of the

dataset. Household-level data typically do not allow one to control for firm characteristics

that can often have important implications for wages and wage inequality (see Meng,

2004 and references therein). A priori, including firm-specific effects should alter the

magnitude of the gender wage gap if (i), the wage gap is correlated, either negatively

or positively, with the firms’ observed and unobserved characteristics; (ii), the wage gap

3A few examples of developing country studies are Diaz et al. (2013) who find that Grit and someof the Big Five traits result in earning gains of 5-10 percent in urban Peru, and Glewwe et al. (2013)who find no evidence that non-cognitive skills matter for wage determination in rural China. But noneof them examine the salience of these traits for gender wage gaps.

3

between males and females is due to gender-based sorting of workers across firms that

pay different wages. For instance, there is evidence of gender segregation across firms

in the African manufacturing sector (Fafchamps et al., 2009) and in small firms in USA

(Carrington and Troske, 1995). If there are high paying firms that hire more men than

women and low paying firms hiring more women, then firms’ characteristics will influence

the gender differences in wages. Controlling for firm heterogeneity should then reduce the

magnitude of the gender wage gap. With linked employer-employee data, we can include

firm-specific effects to account for such firm-level influences on the gender wage gaps.4 We

also use information on firm characteristics to examine correlates of within-firm gender

wage gaps, relying on a two-step procedure including wage regressions and difference in

firm fixed effects regressions.

Our results show that cognitive skills matter more than personality traits in deter-

mining mean wages. Where the personality traits do matter, it is mostly for wages of

female employees, and only in certain parts of the wage distribution. Cognitive skills

as measured by reading and numeracy also seem to confer benefits to women and men

respectively, with returns varying across the wage distribution. The quantile decomposi-

tions indicate that cognitive skills and personality traits reduce the unexplained gender

gap in the upper part of the wage distribution. Finally, the within-firm regressions sug-

gest that employers may make assumptions about the existence of sex-specific skills of

their workers, which could then widen the within-firm gender wage gap.

The paper is organized as follows: the next section discusses the methodology. Section

3 describes the data. Section 4 presents the descriptive statistics and results for mean

and quantile decompositions, and within-firm gender wage gaps. Finally, Section 5 offers

concluding comments.

2 Methodology

2.1 Blinder-Oaxaca Decomposition Framework

We first use the Blinder-Oaxaca method to decompose the mean wage gap between males

and females into portions attributable to differences in the distribution of endowments

(the explained component) and differences in returns to these endowments (the unex-

plained component) (Blinder, 1973; Oaxaca, 1973). This methodology involves estimat-

ing Mincerian wage equations separately for males and females. The decomposition is as

follows:

wm − wf = (Xm − Xf )βm + Xf (βm − βf ) (1)

4A caveat remains that employer-employee data are not representative of the population of interest atthe country level, but to the extent that the firms’ characteristics matter in the wage formation process,inclusion of firm-specific effects yields important advantages in studying wage gaps.

4

where the left hand side of the equation is the difference in the mean log hourly wages

of males and females. Xm and Xf are average characteristics for males and females

respectively and βm and βf are the coefficient estimates from gender-specific OLS regres-

sions. The first term on the right hand side represents the part of the wage differential

due to differences in characteristics and the second term represents differences due to

varying returns to the same characteristics.

In this paper, we also rely on the general decomposition proposed by Neumark (1988)

in which the non-discriminatory wage structure is based on OLS estimates from a pooled

regression (of both males and females) as follows:

wm − wf = (Xm − Xf )β∗ + [(βm − β∗)Xm + (β∗ − βf )Xf ] (2)

Neumark shows that β∗ can be estimated using the weighted average of the wage

structures of males and females and advocates using the pooled sample. The first term

is the gender wage gap attributable to differences in characteristics. The second and the

third terms capture the difference between the actual and pooled returns for men and

women, respectively.

2.2 Quantile Decomposition Framework

Generalising the traditional Blinder-Oaxaca decomposition that decomposes the wage gap

at the mean, Machado and Mata (2005) proposed a decomposition method that involves

estimating quantile regressions separately for males and females and then constructing a

counterfactual using covariates of one group and returns to those covariates for the other

group.

The conditional wage distribution is estimated by quantile regression. The conditional

quantile function Qθ(w|X) can be expressed using a linear specification for each group

as follows:

Qθ(wg|Xg) = XTi,gβg,θ for each θ ∈ (0, 1) (3)

where g = (m, f) represents the groups, w denotes the log of hourly wage, Xi represents

the set of covariates for each individual i and βθ are the coefficient vectors that need to

be estimated for the different θth quantiles.

The quantile regression coefficients can be interpreted as the returns to various char-

acteristics at different quantiles of the conditional wage distribution. Machado and Mata

(2005) estimate the counterfactual unconditional wage distribution using a simulation-

based technique.

Melly (2006) proposed an alternative to the simulation-based estimator that is less

computationally intensive and faster. Instead of using a random sample with replace-

5

ment, Melly (2006) integrates the conditional wage distribution over the entire range of

covariates to generate the marginal unconditional distribution of log wage. Then, by

inverting the unconditional distribution function, the unconditional quantiles of interest

can be obtained. This procedure uses all the information contained in the covariates and

makes the estimator more efficient. Melly (2006) shows that this procedure is numer-

ically identical to the Machado and Mata decomposition method when the number of

simulations used in the latter goes to infinity.

We construct a counterfactual for females using the characteristics of females and the

wage structure for males:

CF fθ = XT

f,iβm,θ (4)

Using the abovementioned counterfactual, the decomposition of wage gaps of the

unconditional quantile function between groups f and m is as follows:

∆θ = (Qm,θ − CF fθ ) + (CF f

θ −Qf,θ) (5)

The first term on the right hand side represents the effect of characteristics (or the

quantile endowment effects) and the second the effect of coefficients (or the quantile

treatment effects).

3 Data

The Bangladesh Enterprise-based Skills Survey (ESS) for 2012 was sponsored by the

World Bank and carried out by a team of the Human Development South Asia Region

(Nomura et al., 2013). The World Bank, together with the government of Bangladesh

and the development partners, had embarked on a comprehensive assessment of the ed-

ucation sector to determine whether it is responding adequately to the skill demands of

firms. The survey contains only formal sector firms. This is a shortcoming of the data as

the Bangladeshi economy heavily leans towards the informal sector. The ESS is a linked

employer-employee survey, containing an employer survey as well as an employee survey

for a random subsample of employees working in the firms surveyed.5 The survey samples

500 firms active in commerce, education, finance, manufacturing and public administra-

tion, while the employee survey samples 6,981 employees. The employer survey, that is

administered to business owners or top managers, consists of a general enterprise pro-

file, including characteristics of the firm and its managers, its recruitment and retention

5From a roster of employees provided by the firm, the random sampling is done as follows: in a smallfirm, every third person was interviewed; in a medium and large firm, every fifth and seventh personswere selected respectively; and if the employment size exceeds 200, every 30th person was interviewed.If a firm did not have a roster, the interviewer listed up all the employees and randomly selected thesamples.

6

practices, and the workforce training it provides. The employee survey contains detailed

information on an employee’s education background, work experience, and household

background information. Further, the employee surveys contain modules to assess cogni-

tive and non-cognitive skills through specific tests. The surveys were conducted between

November 2012 and January 2013 through face-to-face interviews.

The Business Registry of 2009, collected by the Bangladesh Bureau of Statistics, was

used as the sampling frame. The Business Registry contains 100,194 enterprises that have

more than 10 employees. The sampling methodology for the ESS is stratified random

sampling, with the strata being economic sector and firm size. The five economic sectors

selected for sampling were commerce (wholesale/retail), education, finance, manufactur-

ing, and public administration. These five sectors occupy 87 percent of formal sector

enterprises and 91 percent of formal sector employment.6 Enterprises were categorized

into three sizes: small (10-20 employees), medium (21-70 employees) and large (71 or

more employees).

Cognitive skills of employees were measured through literacy and numeracy tests, both

consisting of eight questions. All the questions assess primary education level of cognitive

skills as many of them are taken from National Student Assessment conducted in 2011 by

the Department of Primary Education. The literacy test includes reading of words and

sentences, comprehension of short paragraphs, grammar, and translation from Bangla to

English. The numeracy test consists of simple mathematical operations, measurement,

and functional mathematics, such as cost calculation. Scores are calculated by assigning

one point for each correct answer and then standardized.

Employees’ personality traits were assessed by administering a battery of questions

- taken from the World Bank Skills Toward Employment and Productivity (STEP) sur-

veys - and asking them to answer on a 4-point scale ranging from “almost always”, to

“almost never”.7 Of these questions, 15 items measure the following five personality

factors or traits, commonly identified as the Big Five: openness to experience, consci-

entiousness, extroversion, agreeableness, and neuroticism. The Big Five or Five-Factor

Model is a broadly accepted taxonomy of personality traits (John and Srivastava, 1999).

Openness to experience is the tendency to be open to new aesthetic, cultural, or intel-

lectual experiences. Conscientiousness refers to a tendency to be organized, responsible

and hard working. Extroversion relates to an outward orientation rather than being re-

served. Agreeableness is related to the tendency to act in a cooperative and unselfish

manner. Neuroticism (opposite of emotional stability) is the tendency to experience un-

pleasant emotions easily, such as anger, anxiety, depression, or vulnerability. Further,

6The selection of economic sectors was made purposively. First, the economic sectors have relativelylarge proportion of firms in the formal economic sector as well as large share of employees. Second, theselected economic sectors are considered to have diversity in educational and skills demand.

7These questions are available in the appendix.

7

two questions were used to measure hostile bias which may be defined as over-attributing

hostile intent to peers’ behaviours, even when the actual intent is benign or ambiguous

(Dodge, 2003). Finally, four questions adapted from the Melbourne Decision-Making

Scale (Mann et al., 1997) were used to assess decision-making, that is, whether individu-

als think about the future consequences of their decisions, and if they consider multiple

options when making decisions. Traits are measured by taking the average standardized

score on items corresponding to each trait.8

Since we are interested in within-firm gender wage gaps, we consider firms where at

least one male and one female employee have been sampled, and where we have complete

information on socio-economic characteristics, cognitive skills, and personality traits for

employees. This leaves us with a sample of 225 firms and 2,150 employees.9 Since

wages and personality traits are measured contemporaneously in the survey, one may be

concerned about reverse causality. However, as we discuss in Section 4.1, the average

employee in our sample is 32 years old, falling in the working-age range, during which

personality traits are most stable and any changes are not economically significant (Cobb-

Clark and Schurer, 2012 and 2013).10

4 Results

4.1 Descriptive Statistics

We begin with descriptive statistics of firm characteristics listed in Table 1. 71 percent of

firms report themselves as being profitable. On average, there are 193 employees per firm

of which 25 percent are females. 32 percent of the sample is made up of small firms (10-

20 employees), while medium (21-70 employees) and large firms (71 or more employees)

account for 30 and 38 percent respectively. 61 percent of top managers in firms have a

post-graduate degree. Only a paltry 4 percent of firms have females in top managerial

positions. 97 percent of firms maintain either formal or informal accounts and 96 percent

8The internal reliability of each trait is determined by calculating the Cronbach alphas using the itemscorresponding to each trait. These are: openness to experience, 0.44; conscientiousness, 0.3; extroversion,0.16; agreeableness, 0.3; emotional stability, 0.36; hostile bias, 0.48; decision making, 0.43. While theseinternal reliability coefficients are lower than the range of 0.6-0.7 that has been mostly found in theliterature using longer personality inventories, note that we have three items per trait on average, andthus these values may be deemed satisfactory. Mueller and Plug (2006) discuss that reliability coefficientsincrease with number of items.

9Since a concern may be that that employees who respond to all questions assessing skills and traitsmay be different from those who do not, we checked for the existence of any selection bias with a Heckmanselection model, relying on measures of survey quality (survey controllers and enumerators identifiers)as exclusion restrictions. We did not find evidence of significant selection effects in the wage regressions(results of the selection models are available from the authors upon request).

10In fact, Almlund et al. (2011) discuss that addressing the potential problem of reverse causality byusing previously measured traits as predictors of later outcomes can lead to errors in variables problemif the traits evolve over time.

8

of firms are registered with the government. These two factors reflect the high level of

formality in the sampling frame of the survey, which is based on the Business Registry.

In terms of industrial sectors, the largest chunk of firms (32 percent) is engaged

in manufacturing. Public administration and education make up 20-22 percent each.

Finance firms constitute 19 percent while commerce represents the remaining 7 percent.

Further, within the manufacturing firms, textiles and wearing apparel are the dominant

activities comprising 37 percent and 27 percent respectively while food products make

up 17 percent.

56 percent of firms are based in Dhaka, the capital city. Chittagong, the second largest

city in Bangladesh has 10 percent of the firms, and Rajshahi and Rangpur account for

8-9 percent of firms respectively.

Moving on to employee characteristics in Table 2, out of 2,150 employees, 420 are

female, thereby constituting 19.5 percent of the employee sample. Males are two years

older than females (Wilcoxon rank-sum test; p− value = 0.00) but there is no significant

difference in the proportion of married males and females. Males have approximately

11 years of education, which is 1 year higher than that of females (p − value = 0.09).

Males also exhibit significantly greater tenure at the current firm (p− value = 0.01) and

experience prior to joining the current firm (p − value = 0.04). Given these differences

in endowments, a priori a higher wage for men is expected. As our data show, the

average hourly wage of males is 51 taka (0.51 euros in 2014) while that of females is

approximately 50 taka, with the difference being statistically significant (p − value =

0.00). A Kolmogorov-Smirnov test also shows that the wage distributions of males and

females (as shown in Figure 1) are significantly different. Note that while this wage gap

may appear small for Bangladesh where gender-based inequalities are large and fairly

persistent, one should bear in mind that our sample is comprised of firms in the formal

sector with at least 10 employees, and self-selection of high ability workers into the formal

sector is a priori greater for women than for men (Kingdon, 1998). However, using

national level household survey data for Bangladesh, Asadullah (2006) finds no evidence

of significant sample selection into paid employment for both sexes.11 Moreover, since

the informal sector is not under consideration here, the wage gap measured here is an

under-estimate of the actual income gap characterizing the labour market in Bangladesh.

Moving on to cognitive skills as measured by reading and numeracy tests, our data

show that men outperform women significantly in terms of the numeracy score (p−value= 0.002) but not in terms of reading score (p − value = 0.33). Finally, in terms of

personality traits, females report significantly higher scores on openness to experience

11Kapsos (2008) documents a gender wage gap of 21 percent using the 2007 Bangladesh OccupationalWage Survey. The nature of the sample in this survey comes closest to ours in that a third of the sampleis derived from firms with more than 10 employees.

9

(p − value = 0.001), agreeableness (p − value = 0.06) and neuroticism (p − value =

0.007), decision-making (p − value = 0.001) and hostile bias (p − value = 0.004) as

compared to men.

Another factor that could explain the wage gap is differences in occupational status

between males and females. While 4.6 percent of males and 2.6 percent of females are in

managerial roles, approximately 19 percent of males and females are in professional posi-

tions. Further, while almost 21 percent of women are in unskilled elementary occupations,

the proportion of men is 13 percent.

4.2 The Mean Gender Wage Gap

We first estimate OLS regressions for the pooled sample of males and females and cluster

the standard errors at the firm level. The dependent variable is the log of the current

hourly wage. We subsequently expand the list of explanatory variables. The first set

consists of socio-economic characteristics such as marital status, years of completed ed-

ucation, years of prior experience and years of tenure, with a quadratic profile for the

last three variables. Quadratic effects of experience and tenure are used to approximate

the concave profile between these variables and wages. We also introduce a dummy vari-

able that is equal to 1 when the worker is a woman and zero otherwise. In the second

set, to measure cognitive skills, we further include standardized scores on the reading

and numeracy tests. In the third set, to measure personality traits, we also include

standardized values of scores on each of the Big Five traits (extroversion, agreeableness,

conscientiousness, openness to experience, and emotional stability), and on hostile bias

and decision-making. Next, in each of these regressions, we can pick up the role of unob-

served firm heterogeneity by introducing firm dummies in the regression. Finally, dummy

variables for occupational status are also added to account for the fact that wages within

the same firm could differ on account of different occupations. If the female dummy

variable partially picks up these occupational effects, it would lead to an over-estimated

gender effect. However, a problem is that occupational assignment may itself be the

result of the employer’s practices and not due to differences in productivity or individual

choice (Albrecht et al., 2003). Further, since we are interested in examining the effect of

cognitive and non-cognitive skills on wages, we may not want to control for occupations

if one believes that individuals’ skills determine their occupational preferences.12

Results are in Table 3. In column 1, we regress the log wage on only the female

dummy and obtain a negative coefficient indicating a significant raw gender wage gap of

11.6 percent. In column 2, upon adding the socio-economic controls, the female coefficient

12Results with occupation controls are reported only for the OLS estimates. In all other estimates,we do not include occupation status for reasons outlined above. Results are available from authors fromrequest.

10

reduces significantly to 5.7 percent. In column 3, once the standardized scores on cognitive

tests are added, the gender wage gap remains virtually unchanged at 5.9 percent. The

reading score is positively linked with wages but the numeracy score is not a significant

determinant. While Glewwe et al. (2013) find that reading skills are not significant when

controlling for educational attainment, in our case we find reading proficiency to matter

even in the presence of years of education. In column 4, we further add the standardized

scores of the personality traits, which leads to a slight drop in the female dummy to

5.5 percent. None of the personality traits are statistically significant suggesting that

there is no incremental effect of these traits on mean wages. In columns 5-7, we augment

each of these regressions by adding the firm dummies. The gender dummy coefficients

are smaller in magnitude, as compared to columns 2-4, but are no longer significant.

This indicates the existence of sorting of male and female workers across firms that

pay different wages. An F-test of joint significance of the firm dummy variables shows

them to be highly significant. This shows that wages are correlated with firm-specific

factors, thereby making it crucial to account for firm-specific effects. In the presence of

firm effects, both reading and numeracy scores have positive and significant coefficients.

In column 8, we add occupation dummy variables, with the omitted category being

managerial jobs. Note that the coefficient on the female dummy is reduced while still

being insignificant. Cognitive test scores remain positive and significant. In results not

reported in the table, we find wages on all occupation groups to be significantly lower

than those of the managerial group. Across relevant columns, tests of joint significance

show that cognitive skills are significant in determining mean wages while personality

traits are not. In terms of other independent variables included in Table 3, we find that

the education-wage relationship follows a convex profile. While this is contrary to the

standard assumption of concave relationship between education and earnings, there is

now ample evidence suggesting increasing returns to education across schooling levels

(e.g., Kingdon and Unni, 2001; Soderbom et al., 2006). Tenure in current firm has

the expected concave relationship with wages. These regressions indicate that while

cognitive skills directly affect wages, personality traits do not matter for pooled mean

wages across the sexes. We then proceed to checking for the existence of specific gender

and distributional effects.

4.3 Quantile Regressions

As can be seen in Figure 2, the magnitude of the unconditional gender wage gap varies

considerably throughout the wage distribution with the highest gaps being observed at

the lower percentiles and the smallest (and negative) wage gaps at the highest percentiles.

This phenomenon of greater wage gaps at the lower end is consistent with the “sticky

floor” parabola that has been observed primarily in developing countries (e.g., Chi and Li,

11

2008; Carrillo et al., 2014). We now estimate quantile regressions to determine how the

magnitude of the gender wage gap changes along the wage distribution once we control

for socio-economic characteristics, cognitive skills, and personality traits. By pooling the

data for males and females in the quantile regression, the assumption is that the returns

to endowments are the same at the various quantiles for men and women. In Table 4,

we estimate pooled quantile regressions for the most inclusive specification, without firm-

specific effects. The coefficient of the female dummy varies across the wage distribution

with gaps being higher at the lower end. The gender wage gap is 11 percent at the 10th

percentile, and 14.1 percent at the 25th percentile and then declining to 8.7 percent at

the median. It further declines to 4.9 percent at the 75th and becomes negligible and

insignificant at the 90th percentiles. The reading score is positive and significant at all

percentiles except the 90th. Among the personality traits, openness to experience and

decision-making are negatively associated with wages at the 10th percentile.

In Table 5, we add the firm-specific effects. In order to conduct fixed effects quantile

regressions, we use the method proposed by Canay (2011). This alternative approach

assumes that the unobserved heterogeneity terms have a pure location shift effect on the

conditional quantiles of the dependent variable. In other words, they are assumed to affect

all quantiles in the same way. We notice that with the inclusion of firm-specific effects,

the gender wage gap is lower at all the estimated percentiles of the wage distribution,

as compared to results of Table 4. There is then evidence of sorting across firms by

workers at all points of the wage distribution. The wage gap at the 90th percentile,

while reversed and now marginally in favor of women is not significant. Both reading

and numeracy scores have higher correlations with wages at the lower percentiles than

higher ones. Agreeableness is positively associated with wages at the 10th, 25th and

50th percentiles, and emotional stability is positively associated with wages at the first

decile. On the contrary, openness to experience is negatively associated with wages at the

10th percentile. Conscientiousness is differentially related with wages at both ends of the

conditional distribution: negatively linked with wages at the 10th percentile, but positively

related to wages at the 90th percentile. Hence, while we do not observe significant effects

of non-cognitive skills on mean wages, the observed effects are more nuanced when one

looks across the conditional wage distribution.

In Table 6, we estimate the gender-specific OLS and quantile regressions with firm-

specific effects.13 The reading score is positively associated with female wages at all

reported percentiles but only at the 25th and 50th percentiles for male wages. On the

other hand, the numeracy score is positively correlated with the wages of men at all

points except the 90th percentile, but is never significant for women. While in the pooled

13Results of gender-specific regressions without firm fixed effects are available from the authors uponrequest.

12

quantile regressions with firm-specific effects in Table 5 we saw that the coefficients on

reading and numeracy scores are positive throughout the distribution, it is now evident

that these results are differentiated by gender. Considering the personality variables,

we observe that conscientiousness is rewarded for women at all percentiles except the

90th. Agreeableness also has positive and mostly increasing returns for women across the

conditional wage distribution. On the other hand, openness to experience has increasingly

negative returns for women at the 50th, 75th and 90th percentiles. Interestingly, we find

that emotional stability is negatively related with women’s wages at the median and the

third quartile, although weakly. On the contrary, for men, the effects of personality traits

are more sporadic and reveal no consistent pattern, indicating that these personality

traits are important determinants for women’s wages but not for men’s wages. Nyhus

and Pons (2005) using Dutch data also find that personality traits matter more for wages

of females than males.

Our results of gender-specific quantile regressions provide some support about the

market rewarding individuals who adhere to societal expectations about gender-appropriate

traits and behaviour. For instance, agreeableness (that indicates unselfish and cooper-

ative behaviour) is a positive quality associated with women especially in a developing

country like Bangladesh where gender roles are well-defined and women may be expected

to be more submissive and accommodating in nature (both at home and at the work-

place), and in our sample, we find that more agreeable women are rewarded monetarily

for this trait.14 Further, the result that numeracy is a skill rewarded for men and literacy

a skill rewarded for women also bolsters the argument of gendered cultural expectations,

particularly in a gender-unequal society such as Bangladesh. Guiso et al. (2008) show

that the gender gap in mathematics scores is higher in more gender-unequal societies and

it becomes smaller in more gender-equal cultures. On the other hand, the reading score

gap is usually in favour of women and becomes smaller in more gender-equal cultures.

This could possibly lead to employers forming beliefs about comparative advantages that

men and women have in numeracy and literacy skills respectively and adhering to those

stereotypes.

4.4 Decomposition Analysis

Table 7 reports results from the mean decomposition that decomposes the average wage

gap into explained and unexplained components. Panel A only includes socio-economic

controls for marital status, education, tenure, prior experience, panel B also includes

the standardized scores for the reading and numeracy tests, and finally in panel C, the

14However, note that this result is in contrast to evidence from developed countries where the traitof agreeableness is generally linked to lower wages, perhaps because agreeable individuals may be “toonice”, extremely cooperative, and reluctant to negotiate their wages (e.g., Nyhus and Pons, 2005; Muellerand Plug, 2006).

13

standardized personality scores are also added. In each of the panels, we report results

using the male wage structure, the female wage structure (Blinder, 1973; Oaxaca, 1973),

and the Neumark pooled model. While columns 2 and 3 report the decomposition results

without firm-specific effects, columns 4 and 5 include the firm-specific effects.

Without the firm-specific effects, we see that across all the three panels, using the

Neumark decomposition, about half of the gap is explained by characteristics with the

rest being unexplained. As we move down from panel A to panels B and C, the explained

component first reduces by 2 percentage points from 51.6 percent to 49.3 percent and

then increases to 53.9 percent. This is in line with the coefficient of the female dummy

variable marginally increasing in Table 3 as we move from including only socioeconomic

characteristics in column 2 to also adding cognitive test scores in column 3. However,

with the inclusion of firm-specific effects, the proportion explained increases significantly

within each of the panels, as expected. Looking across panels, 74.8 percent of the wage

gap is explained with only the socio-economic characteristics, and increases to 78.4 per-

cent and reduces negligibly to 78.2 percent upon successively adding cognitive skills and

personality traits respectively. Hence, in the presence of firm-specific effects, controlling

for cognitive and non-cognitive skills increases the mean explained component by about

3.6 percentage points. Mirroring the OLS results from Table 3, upon comparing pan-

els B and C, we find that non-cognitive skills do not have an additional contribution

over and above the cognitive skills in explaining the average gender wage gap. To the

extent that the unexplained component is indicative of gender wage discrimination, con-

trolling for previously unmeasured cognitive and non-cognitive skills gets us to a closer

approximation of actual wage discrimination.

Next, we move to the quantile decompositions performed at the 10th, 25th, 50th, 75th

and 90th percentiles of the distribution. In Tables 8 and 9, we report results using the

male coefficients, i.e. if females were paid like males, without and with firm-specific

effects respectively. Within each of the three panels, it can be seen that the raw wage

gap declines as one moves from the 10th percentile to the 90th percentile. Further, the

share of coefficients declines as one moves to the upper end of the distribution, thereby

supporting the evidence of a sticky floor. This is reflected in the increasing proportion

of the wage gap that can be attributed to differences in characteristics as one moves to

the higher quantiles. In fact, in each of the panels of Table 9, note that at the 75th

and 90th percentiles, differences in characteristics between the sexes (over)explain the

entire wage gap. In panel A of Table 9, the characteristics account for 28 percent of the

wage gap at the 25th percentile, and 123 percent at the 75th percentile; in panel C, the

respective proportions are 24 and 139 percent. Besides, cognitive skills and personality

traits mostly explain the gender wage gaps in the upper part of the conditional wage

distribution as illustrated by the visible increase in the explained component at the 75th

14

and 90th percentiles, as one moves from panel A to panels B and C.

To sum, the decompositions highlight that, contrary to what gender-pooled regressions

on mean wages predict, the effects of cognitive and non-cognitive skills are essentially

gender specific, and are thereby able to explain a non-negligible proportion of the gender

wage gap. Besides, looking at distributional effects, these skills and traits appear to reduce

the unexplained gender gap in the upper part of the conditional wage distribution.

4.5 Determinants of The Within-Firm Gender Wage Gap

In this section, we look at factors on account of which firms pay males and females

differently. In relation to the previous sections, the purpose of this analysis is to identify

how, in particular, employers’ valuation of employees’ cognitive and non-cognitive skills

may affect the magnitude of the within-firm gender wage gap. Indeed, to the extent

that employers value certain skills more than others, and have some common beliefs or

stereotypes about the ability of male and female employees in the absence of perfect

information on their productivity, this could affect the premium they pay to the two

groups of employees.

In order to examine this, we follow a hierarchical modelling approach (Bryk and

Raudenbush, 1992), also applied in Meng (2004) and Nordman and Wolff (2009a), where

wage equations for males and females are first estimated separately using a fixed effects

model:

wmij = βmXmij + θmj + εmij (6)

wfij = βfXfij + θfj + εfij (7)

The male and female firm fixed effects (θm and θf ) retrieved from these regressions re-

flect a premium paid by the firm to its male and female employees respectively, since other

socio-economic characteristics (marital status, years of education, prior work experience,

and tenure at current firm) have already been controlled for in the above regressions.15

The difference between the male and female firm fixed effects (θm - θf ) is an estimate

of the within-firm gender wage gap. For this exercise, from the original sample of 500

firms and 6,981 employees, we restrict the sample to those firms that have at least two

male and two female employees.16 This leaves us with a sample of 165 firms and 3,231

employees (2,494 males and 737 females).

In a second step, we introduce a host of firm-level characteristics in order to explain

15Our results are robust to controlling for occupations in the first stage regressions.16If there is only one person of each sex in a firm, the estimated firm effect would be equal to the

residual estimated for this person and firm and individual residuals cannot be separated.

15

this within-firm wage gap and use OLS regressions. The firm level characteristics we

consider are: economic sector dummy variables, size of the firm, whether the firm is

reported profitable, export status17, age of the firm, proportion of female employees in

the overall firm workforce, proportion of females in top managerial roles, whether the

firm conducts a performance review from time to time18, whether the manager is female

and whether the manager has completed college and higher levels of education.

Employers are also asked to state on a scale of 1-10 (with 10 being most impor-

tant) how important they think it is for employees, both managers/professionals and non-

professionals, to have each of the following skills: communication, team work, problem-

solving, literacy and numeracy, customer care, responsibility, reliability and trustworthi-

ness, creativity, and vocational job-specific skills.19 We use the responses on each of these

by computing standardized scores. In addition, employers are asked to list the top 3 cog-

nitive and non-cognitive skills (among 14 skills in the questionnaire) they use to make

decisions regarding compensation and promotions, separately for managers/professionals

and non-professionals.20 We then construct dummy variables for professionals and non-

professionals respectively taking the value 1 if employers declared any non-cognitive skills

to be either first or second most important, as compared to cognitive skills.

In column 1 of Table 10, we report the estimates of the within-firm gender wage gap

including a set of firm characteristics and two dummy variables for whether employers

value non-cognitive skills more than cognitive skills for compensation and promotion

decisions for professionals and non-professionals respectively. As is evident, this variable

is positive and significant for non-professionals only, implying that employers who place

more weight on non-cognitive skills for this group of workers have larger gender wage

gaps within their firms, that works to the disadvantage of female employees. This might

be due to the fact that these employers use non-cognitive skills as signals in the absence

of better information on cognitive credentials, in particular for women.

In columns 2 and 3, we introduce standardized scores of importance given by employ-

17A large literature in economic theory suggests that a firm’s industrial affiliation, profitability, sizeand openness to international trade should all affect its compensation policy.

18Some studies indicate that firms with less information about individuals’ productivity may pay onegroup of workers (males) more than the other group (females), the latter group conveying potentiallyless reliable information about its productivity (Lundberg and Startz, 1983). Hence, the gender wagegap may result from a lack of information. To control for this, it is then useful to identify firms forwhich individual productivity can be measured. We proxy this with a dummy taking value 1 if the firmconducts a performance review from time to time.

19Professional staff includes managers; professionals; and technical and associate professionals. Non-professional staff refers to clerical support workers; service workers; sales workers; skilled agricultural,forestry and fishery workers; construction, craft and related trade workers; plant and machine operators,assemblers, drivers; and those in elementary occupations.

20The set of cognitive skills are: problem solving, literacy, numeracy, ICT, general vocational job-specific skills, advanced vocational job-specific skills and English language, while the non-cognitive skillsinclude communication, team work, customer care, responsibility, reliability and trustworthiness, moti-vation and commitment, creativity, and confidence.

16

ers to a number of cognitive and non-cognitive skills listed above, for non-professionals

(column 2) and for professionals (column 3) to save on degrees of freedom. While none

of the coefficients are significant for professionals, we find that firms that value problem-

solving skills for non-professional workers have significantly greater gender wage gaps.

This effect is robust when one considers a regression including all the regressors for pro-

fessionals and non-professionals (column 4). Finally, in the most inclusive specification

of column 4, valuing communication skills among non-professional workers appears to be

negatively associated with the within-firm gender wage gap. Hence, in the absence of per-

fect observation of such skills among employees, perhaps employers make the assumption

that males are better endowed than females with certain skills (such as problem solving),

which would tend to increase the gender gap in the wage premium. In the same way, a

supposed feminine trait like communication skills tends to decrease such a gap.21

In terms of other firm characteristics, the share of females in the top management and

the top manager being female is associated with smaller wage gaps within firms. This

effect is robust across all specifications reported. Other variables such as education level

of the manager, size of the firm, sector of operation, and proportion of females in the

overall workforce are not consistently significant across columns in explaining within-firm

gender wage gaps.

5 Discussion and Conclusion

In this paper, our objective has been to explain gender wage gaps in the formal sector of

Bangladesh by including measures of cognitive skills and personality traits as additional

determinants of wages. We believe it makes an important contribution especially when

the existing literature on these issues is scarce for developing countries.

Our results show that, for the particular sample at hand, while cognitive skills, as

measured by reading ability, are positively correlated with wages, measures of personality

seem to have almost no explanatory power in determining mean wages for the pooled

sample. Where the personality traits do matter, it is mostly for wages of female employees,

and in certain parts of the wage distribution. Interestingly, the finding that reading and

numeracy skills are positively correlated with wages across the distribution is driven by

employees’ sex such that reading and numeracy skills seem to confer benefits to men and

women, respectively. Further, in line with OLS regressions, mean decompositions indicate

that including measures of cognitive skills reduces the unexplained component by 3.6

percentage points when firm effects are also accounted for, but personality traits do not

21Alternative modeling, i.e. including cognitive and personality traits in the first step wage regressions(thus reducing the firm sample size to 108 observations due to missing observations on these traits),provides additional evidence that team-work skills tends to reduce significantly the within-firm genderwage gap.

17

play a role. Consistent with other developing country literature, quantile decompositions

indicate the presence of a sticky floor phenomenon, which is revealed by higher adjusted

wage gaps at the lower end of the conditional wage distribution, with the unexplained

component being larger at lower percentiles. Cognitive skills and personality traits greatly

reduce the unexplained gender gap in the upper part of the conditional wage distribution.

The finding that cognitive skills and socio-economic characteristics are more impor-

tant, in general, than personality traits in determining wages is also supported by the

outlook of employers in our sample. In the data, employers are asked to rate how im-

portant the following criteria are on a 1-10 scale when making hiring decisions (10 being

very important): academic performance, work experience, job skills and interview. 68

percent, 57 percent and 50 percent of employers rated academic performance, work ex-

perience, job skills respectively between 8 and 10. On the other hand, only 36 percent

of employers considered interview to be an important selection criteria. This suggests

that employers place greater consideration on observables such as academic performance

and prior work experience, rather than on a face-to-face interaction during an interview,

which gives them the opportunity to assess certain soft skills of the person such as their

assertiveness, agreeableness, communication skills, etc. This could also point to low lev-

els of learning and cognition in countries such as Bangladesh, which might explain why

employers care more about evaluating those skills. These results are in contrast to studies

based on developed countries such as the United States that find employers rank “atti-

tude” as the most important skill among new hires (Bowles et al., 2001). Our results

are however in line with other developing country studies such as Glewwe et al. (2013)

that find cognitive skills to matter more than non-cognitive skills for determining wages

in rural China.

Finally, we also have investigated the determinants of the within-firm gender wage

gap. Share of top position females in the firm and sex of the manager all seem to be

robust significant determinants of the wage gap observed inside the firm. Besides, in the

absence of perfect observation of workers’ productivity and skills as hypothesized above,

employers appear to rely on signals to set wages. These signals may be based on skill

preference and beliefs in the existence of sex-specific skills. For instance, employers seem

to better remunerate those workers making proof of typical perceived sex-specific skills.

How and why such stereotypes persist and cause gender inequality in labour market

outcomes in Bangladesh (and more generally in developing countries) would be worth

investigating further.

18

References:

Ahmed, S., and Maitra, P. (2010). Gender wage discrimination in rural and urban

labour markets of Bangladesh. Oxford Development Studies, 38(1), 83-112.

Ahmed, S., and Maitra, P. (2011). A Distributional Analysis of the Gender Wage

Gap in Bangladesh. Monash University Working Paper.

Albrecht, J., Bjorklund, A., and Vroman, S. (2003). Is there a glass ceiling in Sweden?

Journal of Labor Economics, 21(1), 145-177.

Almlund, M., Duckworth, A.L., Heckman, J.J., and Kautz, T.D. (2011). Personality

Psychology and Economics in Eric A. Hanushek, Stephen Machin and Ludger Woessmann

(Eds.) Handbook of Economics of Education, Vol.4, 1-181.

Appleton, S., J. Hoddinott, and P. Krishnan. (1999). The Gender Wage Gap in Three

African Countries. Economic Development and Cultural Change, 47(2), 289-312.

Asadullah, M. N. (2006). Returns to Education in Bangladesh. Education Economics,

14(4), 453-468.

Babcock, L., and Laschever, S., 2003. Women Don’t Ask: Negotiation and the Gender

Divide. Princeton University Press, Princeton NJ

Bangladesh Bureau of Statistics (2011). Report on Labour Force Survey 2010

Blinder, A. (1973). Wage discrimination: Reduced form and structural estimates.

Journal of Human Resources, 8(4), 436-455.

Borghans, L., Duckworth, A.L., Heckman, J.J., and ter Weel, B. (2008). The Eco-

nomics and Psychology of Personality Traits. Journal of Human Resources, 43(4), 972-

1059.

Bowles, S., Gintis, H., and Osborne, M. (2001). Incentive-enhancing preferences:

Personality behavior and earnings. American Economic Review Papers and Proceedings,

91, 155-158.

Braakmann, N. (2009). Psychological Traits and the Gender Gap in Full-Time Em-

ployment and Wages: Evidence from Germany. Working Paper.

Bryk, A., and Raudenbush, S. W. (1992). Hierarchical Linear Models for Social and

Behavioral Research: Applications and Data Analysis Methods. Newbury Park, CA: Sage.

Canay, I. A. (2011). A simple approach to quantile regression for panel data. The

Econometrics Journal, 14(3), 368-386.

Card, D., Cardoso, A.R., and Kline, P. (2013). Bargaining and the Gender Wage

Gap: A Direct Assessment. IZA Discussion Paper # 7592.

Carrillo, P., Gandelman, N., and Robano, V. (2014). Sticky floors and glass ceilings

in Latin America. Journal of Economic Inequality, 12(3), 339-361.

19

Carrington, William J. and Kenneth R. Troske. (1995). Gender Segregation in Small

Firms. Journal of Human Resources, 30 (3), 503-533.

Chi, W., and Li, B. (2008). Glass ceiling or sticky floor? Examining the gender

earnings differential across the earnings distribution in urban China, 1987-2004. Journal

of Comparative Economics, 36(2), 243-263.

Cobb-Clark, D.A., and Schurer, S. (2012). The stability of Big-Five personality traits.

Economics Letters, 115(1), 11-15.

Cobb-Clark, D.A., and Schurer, S. (2013). Two economists’ musings on the stability

of locus of control. Economic Journal, 123, F358-F400.

Cobb-Clark, D.A., and Tan, M. (2011). Noncognitive skills, occupational attainment,

and relative wages. Labour Economics, 18(1), 1-13.

Croson, R., and Gneezy, U. (2009). Gender differences in preferences. Journal of

Economic Literature, 47(2), 448-74.

Diaz, J.J., Arias, O., and Tudela, D.V. (2013). Does perseverance pay as much as

being smart? The returns to cognitive and non-cognitive skills in urban Peru. Working

paper.

Dodge, K. A. (2003). Do Social Information Processing Patterns Mediate Aggressive

Behavior? In B.B. Lahey, T.E. Moffitt, and A. Caspi (Eds.) Causes of Conduct Disorder

and Juvenile Delinquency. New York: Guilford Press.

Fafchamps, M., M. Soderbom, and N. Benhassine. (2006). Wage Gaps and Job

Sorting in African Manufacturing. Journal of African Economies, 18(5), 824-868.

Flory, J. A., A. Leibbrandt and J. A. List (2015). Do Competitive Workplaces De-

ter Female Workers? A Large-Scale Natural Field Experiment on Job Entry Decisions.

Review of Economic Studies, 82, 122-155.

Fortin, N. (2008). The Gender Wage Gap among Young Adults in the United States:

The Importance of Money vs. People. Journal of Human Resources, 43, 886-920.

Guiso, L., Monte, F., Sapienza, P., and Zingales, L. (2008). Culture, Gender, and

Math. Science, 320, 1164-1165.

Glewwe, P., Huang, Q., and Park, A. (2013). Cognitive Skills, Noncognitive Skills,

and the Employment and Wages of Young Adults in Rural China. Working Paper.

Heath, R., and Mobarak, M. (2015). Manufacturing growth and the lives of Bangladeshi

women. Journal of Development Economics, 115, 1-15.

Heckman, J.J., Humphries, J.E., Urzua, S., and Veramendi, G. (2011). The ef-

fects of educational choices on labor market, health, and social outcomes. Unpublished

manuscript, University of Chicago, Department of Economics.

20

Heckman, J.J., Stixrud, J., and Urzua, S. (2006). The effects of cognitive and noncog-

nitive abilities on labor market outcomes and social behavior. Journal of Labor Eco-

nomics, 24 (3), 411-482.

Jellal, M., Nordman, C.J., and Wolff, F-C. (2008). Evidence on the glass ceiling effect

in France using matched worker-firm data. Applied Economics, 40, 3233-3250.

John, O. P., and Srivastava, S. (1999). The Big-Five trait taxonomy: History, mea-

surement, and theoretical perspectives. In L.A. Pervin and O. P. John (Eds.), Handbook

of personality: Theory and research, Vol. 2, 102-138.

Kapsos, S. (2008). Changes in Employment in Bangladesh, 2000-2005: The Impacts

on Poverty and Gender Equity. Working paper series, ILO Asia-Pacific.

Khanna, S. (2012). Gender wage discrimination in India: Glass ceiling or sticky floor?

CDE Working Paper #214, Delhi School of Economics.

Kingdon, G., and Unni, J. (2001). Education and Women’s Labour Market Outcomes

in India. Education Economics, 9(2), 173-94.

Kingdon, G. (1998). Does the Labour Market Explain Lower Female Schooling in

India? Journal of Development Studies, 35(1), 39-65.

Lindqvist, E., and Vestman, R. (2011). The Labor Market Returns to Cognitive

and Noncognitive Ability: Evidence from the Swedish Enlistment. American Economic

Journal: Applied Economics, 3(1), 101-128.

Lundberg ,S.J., and Startz, R. (1983). Private Discrimination and Social Intervention

in Competitive Labor Markets. American Economic Review, 73, 340-347.

Machado, J., and Mata, J. (2005). Counterfactual Decomposition of Changes in Wage

Distributions using Quantile Regression. Journal of Applied Econometrics, 20, 445-465.

Mann, L., Burnett, P., Radford, M., and Ford, S. (1997). The Melbourne Decision

Making Questionnaire: An instrument for measuring patterns for coping with decisional

conflict. Journal of Behavioral Decision Making, 10(1), 1-19.

Melly, B. (2006). Estimation of counterfactual distributions using quantile regression.

Review of Labor Economics, 68, 543-572.

Meng, X. (2004). Gender earnings gap: the role of firm specific effects. Labour

Economics, 11, 555-573.

Mueller, G., and Plug, E. (2006). Estimating the effects of personality on male and

female earnings. Industrial and Labor Relations Review, 60 (1), 3-22.

Neumark, D. (1988). Employers Discriminatory Behaviour and the Estimation of

Wage Discrimination. Journal of Human Resources, 23(3), 279-295.

Niederle, M. and Vesterlund, L. (2007). Do Women Shy Away from Competition? Do

Men Compete to Much? Quarterly Journal of Economics, 122(3), 1067-1101.

21

Nomura, S., S.Y. Hong, C.J. Nordman, L.R.Sarr, L. and A.Y. Vawda (2013). An

assessment of skills in the formal sector labor market in Bangladesh : a technical report

on the enterprise-based skills survey 2012. South Asia Human Development Sector report

No. 64. Washington DC; The World Bank.

Nordman, C.J., and Roubaud, F. (2009). Reassessing the gender wage gap in Mada-

gascar: does labour force attachment really matter? Economic Development and Cultural

Change, 57(4), 785-808.

Nordman, C. J., A.S. Robilliard, and F. Roubaud. (2011). Gender and ethnic earnings

gaps in seven West African cities. Labour Economics, 18, Supplement 1, pp. S132-S145.

Nordman, C.J., and Wolff F-C. (2009a). Is There a Glass Ceiling in Morocco? Evi-

dence from Matched Worker-Firm Data. Journal of African Economies, 18(4), 592-633.

Nordman, C.J., and Wolff F-C. (2009b). Islands Through the Glass Ceiling? Evidence

of Gender Wage Gaps in Madagascar and Mauritius. In Labor Markets and Economic

Development, Kanbur R. and Svejnar J. (eds), Chapter 25, pp. 521-544, Routledge

Studies in Development Economics, Routledge.

Nyhus, E., and Pons, E. (2005). The effects of personality on earnings. Journal of

Economic Psychology, 26(3), 363-384.

Oaxaca, R. (1973). Male-female wage differentials in urban labor markets. Interna-

tional Economic Review, 14(3), 693-709.

Pham, T., and Reilly, B. (2007). The gender pay gap in Vietnam: 1993-2002: A

quantile regression approach. Journal of Asian Economics, 18(5), 775-808.

Semykina, A., and Linz, S.J. (2007). Gender differences in personality and earnings:

evidence from Russia. Journal of Economic Psychology, 28, 387-410.

Soderbom, M., F. Teal, A. Wambugu and G. Kahyarara (2006). The Dynamics of

Returns to Education in Kenyan and Tanzanian Manufacturing. Oxford Bulletin of Eco-

nomics and Statistics, 68(3), 261-88.

22

Table 1: Descriptive Statistics of Firm Characteristics

Variable Mean SDMaking profit 0.71 0.45Number of employees 192.8 810.8Share of female employees 0.25 0.17Top manager: female 0.04 0.19Top manager: post-graduate level education 0.61 0.49Small (10-20 employees) 0.324 0.47Medium (21-70 employees) 0.298 0.46Large (71+ employees) 0.378 0.49Maintain accounts (either formal or informal) 0.97 0.16Registered with government 0.96 0.21Industrial sector:Commerce 0.067 0.25Education 0.227 0.42Finance 0.191 0.39Manufacturing 0.316 0.46Public Admn 0.2 0.4Location:Rajshahi 0.093 0.29Khulna 0.071 0.26Dhaka 0.56 0.50Chittagong 0.102 0.30Barisal 0.049 0.22Sylhet 0.044 0.21Rangpur 0.08 0.27Number of firms 225

Source: Authors’ calculations using the Bangladesh Enterprise Skills Survey, 2012.

23

Table 2: Descriptive Statistics of Employee Characteristics

Variable All Males FemalesFemales 0.195

(0.39)Hourly wage (in taka) 50.91 50.92 50.87

(54.75) (45.76) (82.06)Ln(hourly wage) 3.69 3.72 3.60

(0.63) (0.61) (0.69)Age 32.26 32.62 30.81

(8.49) (8.51) (8.27)Married 0.80 0.80 0.80

(0.40) (0.40) (0.40)Years of education 10.57 10.69 10.10

(4.92) (4.85) (5.21)Tenure in current firm 5.94 6.09 5.3

(6.08) (6.17) (5.66)Years of prior experience 1.92 1.98 1.65

(2.93) (3.01) (2.58)Cognitive Skills:Reading test score 4.96 5.01 4.77

(2.54) (2.49) (2.75)Numeracy test score 5.76 5.84 5.44

(1.99) (1.94) (2.17)Personality Traits:Openness to experience 2.60 2.59 2.68

(0.59) (0.59) (0.59)Conscientiousness 2.86 2.87 2.83

(0.58) (0.57) (0.58)Extroversion 2.32 2.32 2.33

(0.49) (0.50) (0.49)Agreeableness 2.57 2.56 2.62

(0.58) (0.57) (0.60)Emotional Stability 2.71 2.73 2.65

(0.56) (0.56) (0.56)Hostile Bias 2.38 2.36 2.47

(0.73) (0.72) (0.74)Decision-making 2.56 2.54 2.64

(0.55) (0.55) (0.54)Number of employees 2150

Source: Authors’ calculations using the Bangladesh EnterpriseSkills Survey, 2012. Standard deviation reported in parentheses.The maximum score for the reading and numeracy tests is 8. Themaximum score for the personality traits is 4.

24

Tab

le3:

OL

SR

egressions

Col.1

Col.2

Col.3

Col.4

Col.5

Col.6

Col.7

Col.8

Fem

ale-0.116

∗∗-0.057

∗∗-0.060

∗∗-0.055

∗-0.038

-0.033-0.033

-0.021(0.051)

(0.029)(0.029)

(0.028)(0.029)

(0.028)(0.028)

(0.027)

Married

0.052∗

0.054∗

0.052∗

0.0300.032

0.0310.024

(0.029)(0.029)

(0.029)(0.029)

(0.028)(0.028)

(0.026)

Years

ofE

ducation

0.008-0.001

-0.0000.003

-0.019∗

-0.019∗

-0.025∗∗

(0.011)(0.012)

(0.011)(0.011)

(0.011)(0.011)

(0.012)

Years

ofE

ducation

squared

/1000.341

∗∗∗0.358

∗∗∗0.353

∗∗∗0.364

∗∗∗0.418

∗∗∗0.417

∗∗∗0.291

∗∗∗

(0.053)(0.052)

(0.051)(0.054)

(0.053)(0.053)

(0.059)

Ten

ure

incu

rrent

firm

0.033∗∗∗

0.032∗∗∗

0.032∗∗∗

0.040∗∗∗

0.039∗∗∗

0.039∗∗∗

0.037∗∗∗

(0.006)(0.006)

(0.006)(0.006)

(0.005)(0.005)

(0.005)

Ten

ure

incu

rrent

firm

squared

/100-0.053

∗∗-0.051

∗∗-0.050

∗∗-0.070

∗∗∗-0.068

∗∗∗-0.069

∗∗∗-0.067

∗∗∗

(0.023)(0.023)

(0.023)(0.021)

(0.021)(0.021)

(0.020)

Prior

Exp

erience

0.0180.018

0.0170.015

0.0130.014

0.007(0.016)

(0.016)(0.016)

(0.016)(0.015)

(0.015)(0.014)

Prior

Exp

erience

squared

/1000.044

0.0460.049

0.0670.074

0.0730.084

(0.073)(0.074)

(0.074)(0.069)

(0.069)(0.068)

(0.067)

Read

ing

Score

0.047∗∗

0.050∗∗

0.052∗∗

0.053∗∗

0.055∗∗∗

(0.021)(0.021)

(0.021)(0.021)

(0.020)

Num

eracyScore

-0.015-0.012

0.040∗

0.039∗

0.037∗∗

(0.020)(0.020)

(0.021)(0.020)

(0.019)

Op

enness

toex

perien

ce0.002

-0.003-0.008

(0.015)(0.014)

(0.013)

Con

scientiou

sness

0.004-0.004

-0.005(0.013)

(0.014)(0.014)

Extroversion

0.005-0.009

-0.013(0.013)

(0.012)(0.012)

Agreeab

leness

0.0060.018

0.017(0.017)

(0.013)(0.013)

Em

otional

Stab

ility0.005

0.0070.012

(0.013)(0.011)

(0.011)

Hostile

Bias

-0.0000.003

0.001(0.012)

(0.012)(0.011)

Decision

Mak

ing

-0.024-0.005

0.001(0.015)

(0.013)(0.013)

Con

stant

3.718∗∗∗

2.917∗∗∗

2.990∗∗∗

2.995∗∗∗

2.934∗∗∗

3.089∗∗∗

3.086∗∗∗

3.735∗∗∗

(0.033)(0.073)

(0.082)(0.080)

(0.072)(0.064)

(0.063)(0.110)

Observation

s2150

21502150

21502150

21502150

2150R

20.005

0.4880.490

0.4910.661

0.6650.666

0.696F

irmeff

ectsN

oN

oN

oN

oY

esY

esY

esY

esO

ccupation

controls

No

No

No

No

No

No

No

Yes

Note:

Dep

end

ent

varia

ble

islo

gof

curren

th

ou

rlyw

age.

Stan

dard

errorsclu

steredat

the

firm

levelare

reported

in

paren

theses.

***

signifi

cant

at1%

,**sign

ifican

tat

5%

,*sign

ifican

tat

10%

.

25

Table 4: Quantile Regressions

Q10 Q25 Q50 Q75 Q90

Female -0.111∗∗∗ -0.141∗∗∗ -0.087∗∗∗ -0.049 -0.007(0.038) (0.040) (0.027) (0.031) (0.048)

Married 0.109∗∗∗ 0.099∗∗∗ 0.057∗∗ 0.049 0.028(0.037) (0.035) (0.025) (0.037) (0.041)

Years of Education -0.017 -0.010 -0.011 -0.008 0.007(0.015) (0.013) (0.009) (0.010) (0.015)

Years of Education squared/100 0.390∗∗∗ 0.361∗∗∗ 0.380∗∗∗ 0.397∗∗∗ 0.418∗∗∗

(0.054) (0.052) (0.038) (0.050) (0.078)

Tenure in current firm 0.011 0.030∗∗∗ 0.032∗∗∗ 0.036∗∗∗ 0.041∗∗∗

(0.008) (0.007) (0.007) (0.007) (0.008)

Tenure in current firm squared/100 -0.005 -0.063∗∗ -0.053∗ -0.044 -0.069∗∗

(0.028) (0.029) (0.027) (0.031) (0.028)

Prior Experience -0.001 0.009 0.030∗∗ 0.028∗ 0.014(0.013) (0.014) (0.013) (0.014) (0.018)

Prior Experience squared/100 0.061 0.033 -0.013 0.007 0.151(0.048) (0.063) (0.071) (0.083) (0.101)

Reading Score 0.060∗∗ 0.055∗∗∗ 0.056∗∗∗ 0.067∗∗∗ 0.030(0.030) (0.020) (0.018) (0.020) (0.025)

Numeracy Score 0.014 0.000 -0.010 -0.028 -0.038(0.020) (0.015) (0.011) (0.019) (0.026)

Openness to experience -0.034∗ -0.007 0.003 0.011 0.005(0.019) (0.013) (0.011) (0.017) (0.017)

Conscientiousness -0.001 -0.014 0.003 0.016 0.021(0.018) (0.013) (0.012) (0.014) (0.019)

Extroversion -0.005 0.016 0.006 0.017 0.000(0.015) (0.012) (0.009) (0.013) (0.014)

Agreeableness 0.014 0.019 0.004 -0.011 -0.000(0.021) (0.014) (0.011) (0.015) (0.015)

Emotional Stability 0.023 0.013 -0.006 -0.006 0.006(0.020) (0.012) (0.012) (0.015) (0.016)

Hostile Bias -0.001 0.003 0.003 0.006 0.002(0.017) (0.013) (0.009) (0.015) (0.019)

Decision Making -0.031∗ -0.014 -0.017 -0.012 -0.026(0.017) (0.015) (0.012) (0.016) (0.017)

Constant 2.720∗∗∗ 2.834∗∗∗ 3.041∗∗∗ 3.209∗∗∗ 3.294∗∗∗

(0.096) (0.085) (0.062) (0.062) (0.081)

Observations 2150 2150 2150 2150 2150

Note: Dependent variable is log of current hourly wage. Bootstrapped standard errors based on 100replications are reported in parentheses. *** significant at 1%,** significant at 5%,* significant at10%.

26

Table 5: Quantile Regressions (with firm specific effects)

Q10 Q25 Q50 Q75 Q90

Female -0.075∗∗∗ -0.073∗∗∗ -0.045∗∗ -0.042∗ 0.009(0.027) (0.023) (0.019) (0.023) (0.036)

Married 0.047∗ 0.025 0.044∗∗∗ 0.046∗∗ 0.035(0.027) (0.021) (0.016) (0.023) (0.035)

Years of Education -0.034∗∗∗ -0.037∗∗∗ -0.022∗∗∗ -0.007 -0.011(0.011) (0.007) (0.005) (0.008) (0.009)

Years of Education squared/100 0.458∗∗∗ 0.489∗∗∗ 0.428∗∗∗ 0.381∗∗∗ 0.423∗∗∗

(0.050) (0.030) (0.023) (0.038) (0.052)

Tenure in current firm 0.037∗∗∗ 0.041∗∗∗ 0.041∗∗∗ 0.037∗∗∗ 0.037∗∗∗

(0.008) (0.003) (0.005) (0.005) (0.006)

Tenure in current firm squared/100 -0.074∗∗ -0.077∗∗∗ -0.075∗∗∗ -0.052∗∗ -0.044∗∗

(0.031) (0.012) (0.019) (0.021) (0.019)

Prior Experience 0.011 0.011 0.008 0.016 0.021(0.016) (0.010) (0.008) (0.012) (0.015)

Prior Experience squared/100 0.029 0.055 0.072 0.080 0.112(0.080) (0.048) (0.050) (0.065) (0.078)

Reading Score 0.070∗∗∗ 0.071∗∗∗ 0.055∗∗∗ 0.036∗∗∗ 0.033∗

(0.018) (0.015) (0.014) (0.013) (0.017)

Numeracy Score 0.046∗∗∗ 0.032∗∗∗ 0.029∗∗∗ 0.027∗∗ 0.034∗∗

(0.017) (0.011) (0.009) (0.012) (0.017)

Openness to experience -0.024∗ -0.010 -0.001 0.012 0.002(0.014) (0.011) (0.009) (0.011) (0.014)

Conscientiousness -0.033∗∗ -0.012 -0.001 -0.003 0.025∗

(0.016) (0.010) (0.008) (0.010) (0.015)

Extroversion -0.009 -0.011 -0.009 0.002 -0.006(0.010) (0.008) (0.009) (0.010) (0.013)

Agreeableness 0.029∗ 0.023∗∗ 0.020∗∗ 0.018 0.005(0.017) (0.011) (0.008) (0.014) (0.015)

Emotional Stability 0.025∗ -0.000 0.011 0.010 -0.006(0.014) (0.010) (0.010) (0.012) (0.014)

Hostile Bias 0.002 0.002 0.009 -0.001 0.004(0.015) (0.011) (0.010) (0.009) (0.016)

Decision Making -0.024 -0.008 -0.007 -0.003 -0.002(0.015) (0.011) (0.009) (0.011) (0.015)

Constant 2.841∗∗∗ 3.004∗∗∗ 3.080∗∗∗ 3.168∗∗∗ 3.351∗∗∗

(0.065) (0.044) (0.033) (0.051) (0.056)

Observations 2150 2150 2150 2150 2150

Note: Dependent variable is log of current hourly wage. Bootstrapped standard errors based on 100replications are reported in parentheses. *** significant at 1%,** significant at 5%,* significant at10%.

27

Tab

le6:

Gen

der-S

pecifi

cO

LS

and

Quan

tileR

egressions

(with

firm

specifi

ceff

ects)

OL

SQ

10Q

25Q

50Q

75Q

90O

LS

Q10

Q25

Q50

Q75

Q90

Males

Males

Males

Males

Males

Males

Fem

alesF

emales

Fem

alesF

emales

Fem

alesF

emales

Married

0.0330.059

0.0330.043

∗∗0.049

∗∗0.045

0.0740.033

0.078∗

0.075∗∗∗

0.073∗∗

0.084(0.030)

(0.036)(0.026)

(0.018)(0.024)

(0.037)(0.083)

(0.063)(0.044)

(0.021)(0.032)

(0.067)

Years

ofed

ucation

-0.016-0.032

∗∗-0.037

∗∗∗-0.019

∗∗∗-0.009

-0.014-0.057

∗-0.066

∗∗∗-0.054

∗∗∗-0.057

∗∗∗-0.053

∗∗∗-0.065

∗∗∗

(0.012)(0.013)

(0.008)(0.004)

(0.009)(0.012)

(0.032)(0.019)

(0.013)(0.007)

(0.011)(0.025)

Years

ofed

ucation

squared

/1000.402

∗∗∗0.446

∗∗∗0.484

∗∗∗0.418

∗∗∗0.393

∗∗∗0.440

∗∗∗0.588

∗∗∗0.660

∗∗∗0.585

∗∗∗0.588

∗∗∗0.549

∗∗∗0.621

∗∗∗

(0.060)(0.054)

(0.034)(0.021)

(0.040)(0.060)

(0.149)(0.089)

(0.057)(0.029)

(0.052)(0.114)

Ten

ure

incu

rrent

firm

0.039∗∗∗

0.037∗∗∗

0.040∗∗∗

0.044∗∗∗

0.033∗∗∗

0.037∗∗∗

0.052∗∗∗

0.045∗∗∗

0.042∗∗∗

0.051∗∗∗

0.050∗∗∗

0.061∗∗∗

(0.006)(0.008)

(0.004)(0.004)

(0.006)(0.007)

(0.018)(0.017)

(0.007)(0.004)

(0.010)(0.019)

Ten

ure

incu

rrent

firm

squared

/100-0.069

∗∗∗-0.074

∗∗-0.078

∗∗∗-0.096

∗∗∗-0.046

∗-0.042

-0.073-0.041

-0.040-0.074

∗∗∗-0.075

-0.086(0.024)

(0.030)(0.012)

(0.018)(0.025)

(0.027)(0.085)

(0.105)(0.038)

(0.011)(0.057)

(0.094)

Prior

Exp

erience

0.0210.008

0.0140.018

∗∗∗0.022

∗0.045

∗∗∗-0.025

0.0130.005

-0.030∗∗∗

-0.038-0.028

(0.014)(0.016)

(0.008)(0.007)

(0.013)(0.013)

(0.079)(0.049)

(0.019)(0.011)

(0.027)(0.061)

Prior

Exp

erience

squared

/1000.030

0.0790.049

0.0350.031

-0.0300.186

-0.0070.012

0.235∗∗∗

0.2500.200

(0.056)(0.069)

(0.039)(0.033)

(0.056)(0.053)

(0.269)(0.315)

(0.127)(0.084)

(0.198)(0.734)

Read

ing

Score

0.0270.057

∗∗0.046

∗∗∗0.017

0.0130.017

0.126∗

0.082∗

0.101∗∗∗

0.126∗∗∗

0.153∗∗∗

0.156∗∗∗

(0.022)(0.025)

(0.018)(0.012)

(0.017)(0.022)

(0.075)(0.047)

(0.033)(0.013)

(0.032)(0.060)

Num

eracyScore

0.048∗

0.063∗∗∗

0.045∗∗∗

0.032∗∗∗

0.036∗∗∗

0.027-0.015

-0.019-0.016

-0.016-0.014

-0.015(0.026)

(0.016)(0.011)

(0.010)(0.013)

(0.020)(0.036)

(0.031)(0.022)

(0.012)(0.020)

(0.049)

Op

enness

toex

perien

ce-0.007

-0.023-0.015

0.0000.014

-0.017-0.041

-0.007-0.023

-0.041∗∗∗

-0.058∗∗

-0.100∗∗∗

(0.014)(0.019)

(0.011)(0.010)

(0.013)(0.017)

(0.055)(0.031)

(0.018)(0.011)

(0.026)(0.038)

Con

scientiou

sness

-0.008-0.032

∗-0.017

∗-0.000

-0.0050.026

∗0.044

0.053∗

0.043∗∗

0.045∗∗∗

0.038∗∗

0.017(0.017)

(0.017)(0.010)

(0.008)(0.013)

(0.014)(0.043)

(0.028)(0.020)

(0.009)(0.016)

(0.030)

Extroversion

-0.007-0.006

-0.019∗

-0.0120.002

-0.008-0.012

0.011-0.041

∗∗-0.013

0.0090.043

(0.013)(0.015)

(0.011)(0.008)

(0.010)(0.013)

(0.039)(0.027)

(0.017)(0.010)

(0.017)(0.027)

Agreeab

leness

0.0080.027

0.024∗∗

0.011-0.003

-0.0120.102

∗∗0.055

∗0.067

∗∗∗0.103

∗∗∗0.100

∗∗∗0.137

∗∗∗

(0.015)(0.020)

(0.012)(0.010)

(0.013)(0.015)

(0.044)(0.032)

(0.019)(0.011)

(0.019)(0.039)

Em

otional

Stab

ility0.006

0.030-0.003

0.0060.012

-0.015-0.018

-0.008-0.009

-0.018∗

-0.035∗

-0.044(0.012)

(0.018)(0.011)

(0.009)(0.011)

(0.013)(0.032)

(0.032)(0.019)

(0.010)(0.019)

(0.042)

Hostile

Bias

0.001-0.009

-0.0030.008

0.0070.004

-0.0020.070

∗∗∗0.030

-0.004-0.027

∗-0.012

(0.013)(0.017)

(0.011)(0.007)

(0.010)(0.018)

(0.037)(0.024)

(0.019)(0.011)

(0.015)(0.034)

Decision

Mak

ing

-0.007-0.038

∗∗-0.013

-0.007-0.010

0.0150.021

-0.0070.003

0.021∗∗

0.0340.029

(0.015)(0.017)

(0.011)(0.009)

(0.016)(0.016)

(0.049)(0.035)

(0.019)(0.009)

(0.021)(0.039)

Con

stant

3.074∗∗∗

2.832∗∗∗

3.017∗∗∗

3.055∗∗∗

3.191∗∗∗

3.314∗∗∗

3.147∗∗∗

2.827∗∗∗

2.977∗∗∗

3.155∗∗∗

3.284∗∗∗

3.407∗∗∗

(0.070)(0.083)

(0.048)(0.035)

(0.061)(0.071)

(0.188)(0.134)

(0.090)(0.045)

(0.088)(0.165)

Observation

s1730

17301730

17301730

1730420

420420

420420

420

Note:

Dep

end

ent

varia

ble

islo

gof

curren

th

ourly

wage.

Bootstra

pp

edstan

dard

errors

based

on

100

replica

tion

sare

reported

inp

arenth

eses.***

signifi

cant

at1%

,**

signifi

cant

at

5%,*

signifi

cant

at

10%

.

28

Table 7: Mean Wage Decomposition

Col. 1 Col. 2 Col. 3 Col. 4 Col. 5Without firm fixed effects With firm fixed effects

Panel A: Total Difference in Difference in Difference in Difference inOnly socio-economic characteristics difference endowments coefficients endowments coefficients

Male non-discriminatory structure 0.116 0.067 0.049 0.015 0.101Female non-discriminatory structure 0.116 0.057 0.059 0.076 0.04Pooled (Neumark) non-discriminatory structure 0.116 0.060 0.056 0.087 0.029% of wage gap (Neumark) 100 51.6 48.4 74.8 25.2

Panel B:Adding cognitive skills

Male non-discriminatory structure 0.116 0.056 0.06 0.034 0.082Female non-discriminatory structure 0.116 0.055 0.061 0.08 0.036Pooled (Neumark) non-discriminatory structure 0.116 0.057 0.059 0.091 0.025% of wage gap (Neumark) 100 49.3 50.7 78.4 21.6

Panel C:Adding personality scores

Male non-discriminatory structure 0.116 0.06 0.056 0.038 0.078Female non-discriminatory structure 0.116 0.061 0.055 0.08 0.036Pooled (Neumark) non-discriminatory structure 0.116 0.062 0.053 0.091 0.025% of wage gap (Neumark) 100 53.9 46.1 78.2 21.8

Note: Panel A includes education, tenure, experience and the squared terms. Panel B further adds standardized scores for cognitive skills. In Panel C,

standardized personality scores are also included.

29

Table 8: Quantile Decompositions of Log Wage Gaps

Col.1 Col.2 Col.3Percentile Difference Characteristics Coefficients

Panel A: Only socio-economic characteristics

10 0.182∗∗∗ 0.036 0.146∗∗∗

(0.018) (0.035) (0.036)25 0.171∗∗∗ 0.045 0.126∗∗∗

(0.013) (0.029) (0.033)50 0.123∗∗∗ 0.055∗ 0.068∗∗∗

(0.014) (0.027) (0.033)75 0.088∗∗∗ 0.085∗ 0.003

(0.018) (0.039) (0.047)90 0.054∗∗ 0.13∗ -0.077

(0.025) (0.065) (0.071)

Panel B: Adding cognitive skills

10 0.194 0.022 0.172∗∗∗

(0.184) (0.069) (0.063)25 0.173∗∗ 0.037 0.136∗∗∗

(0.08) (0.032) (0.031)50 0.118∗∗∗ 0.055 0.064∗

(0.022) (0.03) (0.035)75 0.082∗∗ 0.077 0.005

(0.042) (0.044) (0.049)90 0.063 0.114 -0.051

(0.132) (0.071) (0.078)

Panel C: Adding personality scores

10 0.187∗∗∗ 0.019 0.168∗∗∗

(0.018) (0.038) (0.034)25 0.167∗∗∗ 0.042 0.125∗∗∗

(0.014) (0.029) (0.029)50 0.111∗∗∗ 0.052 0.059∗

(0.016) (0.03) (0.036)75 0.091∗∗∗ 0.084∗ 0.007

(0.021) (0.037) (0.044)90 0.063∗∗∗ 0.107∗ -0.044

(0.027) (0.055) (0.059)

Note: Bootstrapped standard errors based on 150 replications reported in parentheses. *** significant at1%,** significant at 5%,* significant at 10%. Panel A includes education, tenure, experience and thesquared terms. Panel B further adds standardized scores for cognitive skills. In Panel C, standardizedpersonality scores are also included.

30

Table 9: Quantile Decompositions of Log Wage Gaps (with firm specific effects)

Col.1 Col.2 Col.3Percentile Difference Characteristics Coefficients

Panel A: Only socio-economic characteristics

10 0.163∗∗∗ 0.028 0.134∗∗∗

(0.014) (0.026) (0.028)25 0.141∗∗∗ 0.039 0.102∗∗∗

(0.013) (0.023) (0.025)50 0.109∗∗∗ 0.058 0.051

(0.015) (0.028) (0.034)75 0.076∗∗∗ 0.094∗∗ -0.018

(0.019) (0.039) (0.041)90 0.071∗∗∗ 0.135∗∗∗ -0.064

(0.02) (0.052) (0.051)

Panel B: Adding cognitive skills

10 0.165∗∗∗ 0.011 0.154∗∗∗

(0.015) (0.029) (0.028)25 0.147∗∗∗ 0.039 0.107∗∗∗

(0.015) (0.026) (0.029)50 0.096∗∗∗ 0.051 0.045

(0.016) (0.03) (0.039)75 0.06∗∗∗ 0.083∗∗ -0.023

(0.019) (0.035) (0.045)90 0.072∗∗∗ 0.129∗∗ -0.057

(0.021) (0.049) (0.054)

Panel C: Adding personality scores

10 0.171∗∗∗ 0.009 0.162∗∗∗

(0.013) (0.029) (0.027)25 0.146∗∗∗ 0.035 0.111∗∗∗

(0.012) (0.027) (0.028)50 0.092∗∗∗ 0.049 0.043

(0.016) (0.028) (0.037)75 0.064∗∗∗ 0.089∗∗ -0.025

(0.021) (0.032) (0.04)90 0.064∗∗∗ 0.126∗∗ -0.061

(0.022) (0.047) (0.059)

Note: Bootstrapped standard errors based on 150 replications reported in parentheses. *** significant at1%,** significant at 5%,* significant at 10%. Panel A includes education, tenure, experience and thesquared terms. Panel B further adds standardized scores for cognitive skills. In Panel C, standardizedpersonality scores are also included.

31

Table 10: Within-Firm Gender Wage Gap

Col.1 Col.2 Col.3 Col.4

Education -0.199∗∗ -0.189∗ -0.189∗ -0.176(0.092) (0.106) (0.110) (0.116)

Finance -0.167∗∗ -0.186∗∗ -0.160 -0.161(0.084) (0.093) (0.097) (0.102)

Manufacturing -0.207 -0.280∗∗ -0.180 -0.243(0.125) (0.138) (0.142) (0.151)

Public Admn -0.104 -0.110 -0.086 -0.092(0.103) (0.118) (0.119) (0.125)

21-70 employees -0.108 -0.077 -0.094 -0.082(0.070) (0.064) (0.072) (0.071)

71 or more employees -0.131∗ -0.076 -0.110 -0.071(0.067) (0.065) (0.073) (0.069)

Age of firm 0.001 0.001 0.001 0.001(0.001) (0.001) (0.001) (0.001)

% of females 0.310∗ 0.290 0.311 0.240(0.182) (0.199) (0.205) (0.211)

% of top mgmt females -0.486∗∗∗ -0.422∗∗ -0.414∗∗ -0.385∗

(0.170) (0.188) (0.183) (0.198)

Formal performance review 0.001 0.005 0.009 0.020(0.045) (0.049) (0.047) (0.050)

Firm is profitable 0.048 0.035 0.049 0.034(0.055) (0.056) (0.054) (0.057)

Firm exports 0.021 0.053 0.043 0.053(0.079) (0.084) (0.081) (0.085)

Female manager -0.187∗∗ -0.192∗∗ -0.202∗∗ -0.190∗

(0.089) (0.094) (0.091) (0.097)

Manager college educated -0.153 -0.184∗ -0.164 -0.180∗

(0.106) (0.103) (0.103) (0.096)

Value non-cognitive skills for non-professionals 0.131∗∗

(0.057)

Value non-cognitive skills for professionals 0.012(0.047)

Importance given for non-professionals:Communication skills -0.044 -0.102∗

(0.038) (0.057)

Team work skills -0.029 -0.015(0.033) (0.035)

Problem solving skills 0.086∗∗ 0.099∗∗

(0.039) (0.049)

Literacy skills -0.029 -0.040(0.045) (0.055)

Numeracy skills -0.011 -0.008(0.035) (0.043)

Customer care skills -0.001 -0.006(0.035) (0.040)

Responsibility 0.018 0.019(0.037) (0.047)

Creativity -0.015 -0.038(0.031) (0.042)

Vocational job-specific skills -0.015 -0.012(0.035) (0.059)

Importance given for professionals:Communication skills 0.002 0.061

(0.031) (0.047)

Team work skills 0.002 0.020(0.034) (0.034)

Problem solving skills -0.018 -0.055(0.034) (0.040)

Literacy skills -0.009 0.007(0.033) (0.041)

Numeracy skills -0.004 0.001(0.033) (0.042)

Customer care skills -0.021 -0.005(0.038) (0.039)

Responsibility 0.022 -0.006(0.033) (0.044)

Creativity 0.045 0.053(0.034) (0.044)

Vocational job-specific skills -0.031 -0.005(0.030) (0.053)

Observations 164 164 164 164R2 0.189 0.218 0.189 0.256

Note: Robust standard errors in parentheses. *** significant at 1%,** significant at 5%,*

significant at 10%. 32

Figure 1: Hourly Wage Distributions by Gender

Figure 2: Gender Wage Gap Distribution

33

A Personality Questionnaire

How do you see yourself?

Response scale: Almost always; Most of the time; Some of the time; Almost never

Agreeableness:

Are you very polite to other people?

Are you generous to other people with your time or money?

Do you forgive other people easily?

Conscientiousness:

When doing a task, are you very careful?

Do you work very well and quickly?

Do you prefer relaxation more than hard work?

Neuroticism:

Are you relaxed during stressful situations?

Do you tend to worry?

Do you get nervous easily?

Extroversion:

Are you talkative?

Are you outgoing and sociable, for example, do you make friends very easily?

Do you like to keep your opinions to yourself?

Openness:

Do you come up with ideas other people haven’t thought of before?

Are you very interested in learning new things?

Do you enjoy beautiful things, like nature, art and music?

Hostile Attribution Bias:

Do people take advantage of you?

Are people mean/not nice to you?

Decision-making:

Do you think about how the things you do will affect you in the future?

Do you think carefully before you make an important decision?

Do you ask for help when you don’t understand something?

Do you think about how the things you do will affect others?

34